Intermediate saved values are sometimes `inf`

Hi Patrick! I've run into an issue over in https://github.com/dynamiqs/dynamiqs/issues/666 when t0=t1 and I try to save intermediate values. It seems to be independent of the stepsize_controller that I use (adaptive or constant). Here is a minimal example using constant steps.

import diffrax as dx
import jax.numpy as jnp

term = dx.ODETerm(lambda t, y, _: y)
y0 = jnp.array([1.0])
ts = jnp.array([0.0, 0.0])
saveat = dx.SaveAt(subs=[dx.SubSaveAt(ts=ts), dx.SubSaveAt(t1=True)])

solution = dx.diffeqsolve(
    term,
    dx.Tsit5(),
    ts[0],
    ts[-1],
    0.1,
    y0,
    saveat=saveat,
)
print(solution.ys[0])  # [[inf] [inf]]
print(solution.ys[1])  # [[1.]]

Reproduce the minimal example with diffrax.diffeqsolve, SaveAt, and SubSaveAt when t0 equals t1. Inspect the intermediate-value saving path and compare it with the final-value path; done means the requested intermediate values are finite and match the expected values.